AIM-AI: an Actionable, Integrated and Multiscale genetic map of Alzheimer's disease via deep learning

项目摘要 针对PAR-19-269《阿尔茨海默病遗传和表型数据的认知系统分析》， 在这项提案中，我们组建了一个跨学科团队，以开发新的和强大的分析方法来 有效解决当前在利用遗传学、组学和神经成像数据方面的挑战 阿尔茨海默病(AD)。我们团队的专业知识涵盖复杂疾病遗传学、功能基因组学和 监管、机器学习/深度学习、面向系统的研究、神经成像、药物信息学、 计算神经科学，临床和翻译科学。人工智能(AI)已经被展示出来 在发现对疾病诊断或病因学至关重要的隐藏特征方面具有强大的能力。然而，仅仅是让 人工智能模型“可解释的”对AD的可解释性没有任何影响，包括分子中详细描述的主要影响 生物学、病理学和神经成像。我们的总体目标是开发和实现一个强大的人工智能框架， 即AIM-AI，用于以一种可操作的、集成的和 多尺度，使遗传因素对后续的病因学研究具有明显的实用价值。为了让我们的 发现可行，我们探索在功能上拦截遗传因素影响的多重组学系统 在特定细胞类型和单个细胞分辨率下。我们将发展集成的、脑部数据驱动的集体 系统，涵盖遗传、表型、多组学、细胞背景、神经成像和知识库信息。 最后，将实施多尺度系统生物学方法来识别遗传、神经成像和 表型变化，结合起来可以更好地解释AD的遗传结构及其认知 拒绝。我们将挖掘AD的功能、细胞、组织和细胞类型特定的特征，以及 神经成像水平，支持更严格的评估和验证，即遗传效应确实在 认知衰退与阿尔茨海默病表型。我们的建议有三个具体目标。目标1：培养深度学习能力 框架，“DeepBrain-AD”，用块状脑组织和单细胞来表征AD的遗传风险 监管基因组学。目标2.通过以下方式确定导致阿尔茨海默病进展的认知能力下降的变量 开发深度学习模型，在联合分析中连接多种模式(成像、临床、基因组学) 框架。目标3.使用多重组学数据评估和验证来自目标1和目标2的遗传变异 举例说明调节其影响的分子系统。总而言之，我们将独特地调查和验证 阿尔茨海默病的遗传变异和其他标记在多组学水平、细胞类型背景和单细胞分辨率上； 并将遗传关联信号与功能调节、蛋白质表达和神经成像环境联系起来； 最后解释了它们在阿尔茨海默病进展所致认知功能下降中的作用。这项工程的圆满完成 将生成一个强大的AIM-AI框架，包括机器学习方法/工具、资源和科学 通过整合组学、深度学习和其他基于系统的方法进行发现，这将是 立即与AD和其他疾病研究社区分享。

Project Summary In response to PAR-19-269 “Cognitive Systems Analysis of Alzheimer's Disease Genetic and Phenotypic Data”, in this proposal we assemble an interdisciplinary team to develop novel and robust analytical approaches to effectively address the current challenges in capitalizing on genetics, omics and neuroimaging data in Alzheimer’s disease (AD). Our team expertise covers complex disease genetics, functional genomics and regulation, machine learning/deep learning, systems-oriented research, neuroimaging, drug informatics, computational neuroscience, and clinical and translational science. Artificial intelligence (AI) has been shown powerful in uncovering hidden features that are critical to disease diagnosis or etiology. However, merely making the AI models “explainable” does nothing for explainability of AD, including major effects detailed in molecular biology, pathology, and neuroimaging. Our overall goal is to develop and implement a robust AI framework, namely AIM-AI, for transforming the genetic catalog of AD in a way that is Actionable, Integrated and Multiscale, so that genetic factors have clear utility for subsequent etiological studies. To make our findings Actionable, we explore multiple-omics systems that functionally intercept the effects of genetic factors at the cell-type-specific and single-cell resolution. We will develop Integrated and brain-data-driven collective systems, covering genetic, phenotypic, multi-omics, cell context, neuroimaging and knowledgebase information. Finally, a Multiscale systems biology approach will be implemented to identify genetic, neuroimaging, and phenotypic changes, which in combination can better explain the genetic architecture of AD and its cognitive decline. We will mine the AD characteristics at functional, cellular, tissue- and cell type-specific, and neuroimaging levels, enabling more rigorous assessment and validation that genetics effects indeed play out in cognitive decline and AD phenotypes. Our proposal has three specific aims. Aim 1: Develop a deep learning framework, “DeepBrain-AD”, to characterize the genetic risk of AD using both bulk brain tissue and single-cell regulatory genomics. Aim 2. Identify variants that account for cognitive decline due to AD progression by developing deep learning models that connect multiple modalities (imaging, clinical, genomics) in a joint analysis framework. Aim 3. Assess and validate the genetic variants from Aims 1 and 2 using multiple omics data to illustrate molecular systems which mediate their effects. In summary, we will uniquely investigate and validate genetic variants and other markers in AD at multi-omics level, at the cell-type context and single-cell resolution; and link the genetic association signals with functional regulation, protein expression, and neuroimaging context; and finally explain their roles in cognitive decline due to AD progression. The successful completion of this project will generate a robust AIM-AI framework, including machine learning methods/tools, resources, and scientific discoveries through integrative omics, deep learning, and other systems-based approaches, which will be immediately shared with AD and other disease research communities.